Financial Services Must Secure AI Agents Before Scaling Them
Fifty-seven percent of financial institutions are implementing or planning to implement AI agents, according to Gartner. But most are stuck on the same problem: how to run these systems safely without exposing customer data.
The shift from experimentation to production has exposed a gap. Banks and insurers have the technology. They lack the data foundations to deploy it securely at scale.
"Almost every conversation I have with a CIO or chief risk officer starts with AI agents, but it quickly turns into a conversation about data foundations, governance and trust," said Rinesh Patel, global head of industry for financial services at Snowflake. "Firms have moved past experimentation and are asking how they put agents into production safely, against data they can audit, explain and stand behind."
Where AI Agents Add Value
The clearest early wins are in high-volume, repetitive work. Anti-money laundering investigations offer one example. Analysts traditionally spend hours pulling transaction data, customer records, sanctions lists and other information from separate systems. AI agents now do that in seconds and generate reports for human approval.
"That shifts banks from looking backward to actively defending against financial crime," Patel said.
Mortgage underwriting shows another use case. AI agents can analyze applications, financial documents, property data and risk policies simultaneously, reducing underwriting timelines from days to minutes. Regulatory reporting at global banks and claims processing at insurers follow the same pattern: faster decisions, real-time risk management, better customer experience.
The underlying shift matters more than individual applications. Banks are moving from fragmented, siloed systems to unified data governance frameworks. Done correctly, this approach should be more secure than legacy environments, not less.
Three Shifts in How Security Works
Security is becoming data-centric rather than perimeter-based. Governance, access controls and policies now travel with the data itself, regardless of where it moves or which agent accesses it. This allows banks to deploy AI across the enterprise without compromising sensitive information.
Zero trust is expanding beyond users to include AI agents. Every query, tool call and decision is verified, scoped and auditable. This means agents can be traced, challenged and explained to regulators-a requirement for any institution operating at scale.
Unified governance frameworks are replacing fragmented approaches. A single framework spanning structured and unstructured data, first- and third-party sources, across every cloud and jurisdiction with consistent policies and auditability eliminates the silos that create security blind spots.
"The perimeter is no longer the network, it's the data itself," Patel said.
What Finance Leaders Should Do Now
Three concrete steps can accelerate secure implementation. First, make unstructured data accessible to AI. Contracts, communications and other documents often sit locked in formats AI cannot process. Unlocking this data removes what Patel calls "the biggest lock" most firms face.
Second, create a semantic strategy. Shared definitions and consistent meanings help AI interpret data the same way across systems and teams.
Third, build a connected foundation of first- and third-party data across clouds, jurisdictions and business lines. This foundation must include governance built in at each step, not bolted on at the end.
The underlying principle is straightforward: data strategy matters more than the AI agent itself. "A governed AI agent can unify and synthesize information in real time, surfacing risks, hedging opportunities, liquidity insights and geopolitical exposure before they become operational issues," Patel said. "But only if the data foundation beneath it is unified, governed and connected. Without that foundation, agents operate in a vacuum. With it, they become the most powerful decision-making infrastructure financial services have ever had."
For finance teams considering AI for Finance and AI Agents & Automation, the message is clear: start with your data, not your tools.
Your membership also unlocks:
